--- title: "If Intelligence Is Abundant, What Is the Moat?" url: "https://atlan.com/context-and-chaos/issue/if-intelligence-is-abundant-what-is-the-moat/" description: "AI intelligence is becoming a commodity. The real enterprise moat is context. Here's why performance = intelligence x context, and what it means for your AI strategy." keywords: "Context Engineering, Enterprise AI, AI Strategy, Intelligence Commoditization" --- > Atlan is hosting Context Conference, bringing together the leaders and builders at the frontier of giving AI the context it needs to understand their business. It runs online on October 28, 2026, from 11:00 AM to 2:00 PM ET. Atlan co-founder Prukalpa Sankar opens and closes the day. Leaders from AstraZeneca, BNY and Verizon share why they invest in context and what they get from it. Registrants get early access to The AI Context Gap, a new study from MIT Technology Review Insights. Register: https://atlan.com/context-conference/ A Context and Chaos Deep Dive by Prukalpa Sankar, Co-founder & Co-CEO of Atlan, published May 7, 2026 (11 min read). Subtitle: Performance is intelligence times context. The labs solved one side. Your enterprise has to solve the other. The issue argues that as model intelligence becomes abundant and cheap, the durable enterprise advantage is organizational context. ## Key takeaways - AI intelligence is commoditizing, with a 300x cost reduction in two years; the durable enterprise moat is organizational context, not model capability. - Performance is intelligence multiplied by context, not added. If context is zero, performance is zero, whichever frontier model you choose. - The three enterprise AI failure modes (connectivity, semantic, institutional knowledge) are all context failures. Upgrading the model fixes none of them. ## The paradox - Models keep topping leaderboards: Gemini 3.1 Pro at 94% on PhD-level science reasoning tests; GPT-5.4 Pro solving one of FrontierMath's Open Problems unsolved by mathematicians. Benchmarks keep getting harder. - Frontier-level inference cost dropped nearly 300x in two years. Gartner predicted that by 2030 LLMs will be up to 100x more cost efficient than 2022 models, and inference on a 1-trillion-parameter model will cost 90% less. - Yet MIT found 95% of generative AI pilots fail, and IDC reported 88% of AI POCs never reach production. - The misunderstanding: model intelligence is treated like an SAT score, a narrow benchmark read as a broad measure of success. The highest scorer is not the best employee without onboarding, mentorship and business context; the "smartest" AI cannot deliver value without the systems, knowledge and feedback loops that provide context. ## The next frontier: contextual intelligence The model race optimizes raw reasoning (pattern recognition, inference, code generation, language comprehension). What matters is whether the AI did the right thing in a specific situation. **Contextual intelligence** is the ability to apply an organization's situational knowledge to an AI's reasoning: organizational definitions, domain rules, historical precedents, user intent, environmental constraints, institutional memory and tacit knowledge. - Performance = intelligence x context. The function is multiplicative: zero context means zero performance. Even the best model cannot decide well without knowing your risk scoring convention or which source is the "source of truth" this week. - High intelligence with the wrong context yields negative performance: more elaborate, persuasive and dangerous errors; it hallucinates more convincingly. - The industry treats intelligence as the hard frontier and context as an implementation detail for customers. The opposite is true: labs are solving reasoning at speed; the contextual basis for reasoning is the unsolved frontier. ## Contextual intelligence isn't new Context is what makes enterprise intelligence possible (like walking needs gravity), an insight from decades of cognitive science: - **Lucy Suchman (1987)**: studying Xerox copier use, found people improvise rather than execute plans; the copier's "intelligent" help failed because it saw only button presses, not the situation. Agentic AI assumes the same scripted execution; real performance needs continuous adaptation. - **James Gibson (1979)**, affordances: a chair affords sitting to a human, not a fish. Usefulness lies in the relationship with the environment; model "intelligence" varies by context, not by model. - **Stuart and Hubert Dreyfus (1980)**, five-stage skill model (novice, advanced beginner, competent, proficient, expert): novices follow context-free rules, experts respond to situations intuitively. The more intelligent a system, the more context-sensitive it should be; research on context rot shows more context is not always better. The AGI race builds "perpetual novices." - **Edwin Hutchins (1990s)**, distributed cognition: on naval vessels no single person knows how to pilot the ship; intelligence is spread across people, tools, maps and procedures. The question becomes "how intelligent is the system?" A brilliant model in a poorly configured system does worse than a mediocre model in a well-configured one. - **Yann LeCun (2026)** left Meta after twelve years to launch AMI Labs, raising $1.03 billion, the largest European seed round ever, to build "world models" that understand physics rather than predict text: a bet that intelligence without contextual representation is a dead end. ## Today's intelligence gap in practice Failed pilots rarely lack reasoning power; they suffer context collapse, in three forms: - **Connectivity failure**: an agent analyzing a customer journey sees "John Doe" in the CRM, "J. Doe" in billing and "User_882" in support logs. It reasons perfectly on what it can see but is blindfolded. - **Semantic failure**: asked for "Quarterly Revenue", the model does not know Sales means bookings and Finance means recognized cash, picks one, and produces a sophisticated but useless report. - **Institutional knowledge failure**: a compliance agent flags a transaction per the written manual, unaware of a memo six months ago granting a permanent exception for that entity type, which lives in a buried PDF and a veteran analyst's memory. Upgrading from GPT-4 to GPT-5 or switching from Claude to Gemini fixes none of these. Failures compound: at 85% success per step, a ten-step workflow succeeds only about 20% of the time. OpenAI, building an internal data agent, needed six layers of context: table usage and schema, human annotations, code-derived definitions, institutional knowledge from Slack and Docs, memory from corrections, and runtime context from live queries. ## What this means for the enterprise No company can build a moat on AI alone, a resource that gets exponentially cheaper and is available to every competitor. Context (your understanding of your own data, semantic definitions and historical "why") cannot be trained into a generic model. "While intelligence converges, context compounds." Three shifts: - **The frontier shifts**: from "how do we build a more powerful model?" to "how do we represent and maintain organizational context?" This calls for a science of **context engineering** that turns an analyst's tacit knowledge into a structured world model a machine can navigate. - **The strategic asset shifts**: the durable moat is the organization's world model. The tenth agent inherits the context built for the first nine, so the marginal cost of deploying AI falls as collective accuracy rises ("investing in context creates compounding alpha"). - **The measure of progress shifts**: from MMLU scores and leaderboards to **situated evaluation**, measuring how the whole system performs a specific task in your environment with your messy data. ## Conclusion: the great inversion The hard and easy problems have switched places. Building the reasoning engine is being solved at scale by a handful of labs. Building the context that makes reasoning useful cannot be solved once and shipped globally; it must be solved for every organization, domain and situation. The winners of the next decade will be those that master situated intelligence: capturing tacit knowledge, resolving semantic conflicts and mapping their data landscapes, not those with the largest compute clusters or priciest model subscriptions. (Cat comic: "Upgrading the Model vs Fixing the Context.") ## Related reads - [OpenAI's Frontier Proves Context Matters. But It Won't Solve It.](https://atlan.com/context-and-chaos/issue/openais-frontier-proves-context-matters-but-it-wont-solve-it/) - [Context Graphs as AI Evaluation Infrastructure](https://atlan.com/context-and-chaos/issue/context-graphs-as-ai-evaluation-infrastructure/) - [Conceptual Modeling Is the Context Engineering Nobody Is Doing](https://atlan.com/context-and-chaos/issue/conceptual-modeling-is-the-context-engineering-nobody-is-doing/) - [BI-Ready Is Not AI-Ready](https://atlan.com/context-and-chaos/issue/bi-ready-is-not-ai-ready/) Context & Chaos is a community newsletter on context engineering, governance, architecture and discovery. All issues: https://atlan.com/context-and-chaos/ / Write for it: https://metadataweekly.substack.com/p/write-for-metadata-weekly